Re: any opinion on the patch "[RFC PATCH] hw/i386/e820: remove legacy reserved entries for e820"? EOM

Re: any opinion on the patch "[RFC PATCH] hw/i386/e820: remove legacy reserved entries for e820"? EOM

[Ani Sinha]

On Fri, Mar 4, 2022 at 3:48 PM Gerd Hoffmann <kraxel@redhat.com> wrote:
>
> On Thu, Mar 03, 2022 at 03:12:51PM +0530, Ani Sinha wrote:
> > On Thu, Mar 3, 2022 at 15:11 Gerd Hoffmann <kraxel@redhat.com> wrote:
> >
> > > Sorry, no.  Noticed the discussions but don't remember the details and
> > > didn't took the time to wade through the code to refresh my memory.
> >
> > Could you please take a look when you get time? You have the most context
> > in this space I believe.
>
> Should indeed not be needed unless you use a stone-aged seabios version.
> But I think you can't simply drop it for live migration compatibility
> reasons.  So you'll need do the compatibility dance and drop it for new
> machine types only.  I doubt the benefits outweigh that effort ..

Igor what do you think?

Re: any opinion on the patch "[RFC PATCH] hw/i386/e820: remove legacy reserved entries for e820"? EOM

[Ani Sinha]

On Fri, Mar 4, 2022 at 4:11 PM Ani Sinha <ani@anisinha.ca> wrote:
>
> On Fri, Mar 4, 2022 at 3:48 PM Gerd Hoffmann <kraxel@redhat.com> wrote:
> >
> > On Thu, Mar 03, 2022 at 03:12:51PM +0530, Ani Sinha wrote:
> > > On Thu, Mar 3, 2022 at 15:11 Gerd Hoffmann <kraxel@redhat.com> wrote:
> > >
> > > > Sorry, no.  Noticed the discussions but don't remember the details and
> > > > didn't took the time to wade through the code to refresh my memory.
> > >
> > > Could you please take a look when you get time? You have the most context
> > > in this space I believe.
> >
> > Should indeed not be needed unless you use a stone-aged seabios version.
> > But I think you can't simply drop it for live migration compatibility
> > reasons.  So you'll need do the compatibility dance and drop it for new
> > machine types only.  I doubt the benefits outweigh that effort ..
>
> Igor what do you think?
Since the static entries are separate from the rom file entries, I
think we are in trouble only if the destination is using an old bios?
Otherwise, the non-existence of the static entries should be simply
ignored right?

Re: any opinion on the patch "[RFC PATCH] hw/i386/e820: remove legacy reserved entries for e820"? EOM

[Igor Mammedov]

On Fri, 4 Mar 2022 16:16:38 +0530
Ani Sinha <ani@anisinha.ca> wrote:

> On Fri, Mar 4, 2022 at 4:11 PM Ani Sinha <ani@anisinha.ca> wrote:
> >
> > On Fri, Mar 4, 2022 at 3:48 PM Gerd Hoffmann <kraxel@redhat.com> wrote:  
> > >
> > > On Thu, Mar 03, 2022 at 03:12:51PM +0530, Ani Sinha wrote:  
> > > > On Thu, Mar 3, 2022 at 15:11 Gerd Hoffmann <kraxel@redhat.com> wrote:
> > > >  
> > > > > Sorry, no.  Noticed the discussions but don't remember the details and
> > > > > didn't took the time to wade through the code to refresh my memory.  
> > > >
> > > > Could you please take a look when you get time? You have the most context
> > > > in this space I believe.  
> > >
> > > Should indeed not be needed unless you use a stone-aged seabios version.
> > > But I think you can't simply drop it for live migration compatibility
> > > reasons.  So you'll need do the compatibility dance and drop it for new
> > > machine types only.  I doubt the benefits outweigh that effort ..

well, it's mostly dead and unused code path that needlessly complicates code
and silently bit-rots for several years. (on both SeaBIOS and QEMU sides)
  
> > Igor what do you think?  
> Since the static entries are separate from the rom file entries, I
> think we are in trouble only if the destination is using an old bios?
> Otherwise, the non-existence of the static entries should be simply
> ignored right?

If you take into account migration from older QEMU, it will migrate
with old SeaBIOS (one that VM was started with on source side), then you
can't just remove interfaces it might access from under its foot.
So compat knobs are necessary.

Considering that SeaBIOS switched to QEMU provided ACPI tables since 1.7,
this machine type can serve as point where compat knob should be.
Newer machine type (though technically possible) and SeaBIOS should
not build its own ACPI tables and not use legacy interfaces necessary
for it.

Also since machine types older that 1.7 are deprecated now,
we can remove them in 2 releases, which rules pre-QEMU-ACPI
SeaBIOS out of the picture. So I'd think about preparing cleanup
for SeaBIOS which removes legacy ACPI tables from it, and merge
that once deprecated machine types are removed on QEMU side.
And after that drop legacy ABI on QEMU side as no longer used.